Abstract:We present a new fully-automatic non-rigid 3D shape registration (morphing) framework comprising (1) a new 3D landmarking and pose normalisation method; (2) an adaptive shape template method to accelerate the convergence of registration algorithms and achieve a better final shape correspondence and (3) a new iterative registration method that combines Iterative Closest Points with Coherent Point Drift (CPD) to achieve a more stable and accurate correspondence establishment than standard CPD. We call this new morphing approach Iterative Coherent Point Drift (ICPD). Our proposed framework is evaluated qualitatively and quantitatively on three datasets and compared with several other methods. The proposed framework is shown to give state-of-the-art performance.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the challenges in non - rigid 3D shape registration (morphing), especially when dealing with face datasets with wide variations. Traditional methods use a single template shape, which leads to problems such as slow convergence speed, inaccurate final shape correspondence, and visible residual errors in significant local parts. Specifically:
1. **Non - rigid 3D shape registration problem**: The goal is to align and deform the source point set to the target point set. When dealing with face/head data, the relative positions of key parts (such as ears, mouths, and noses) vary greatly among different individuals, especially in datasets that span a wide range of ages, genders, and races.
2. **Limitations of existing methods**: Current methods usually use the same template shape to deform all members of a specific 3D object class. However, this method performs poorly when dealing with datasets of global object classes because these datasets have wide variations in the spatial distribution of their components. Using a single template shape will cause the key parts of the template to be inconsistent with the relative positions of the key parts of the original 3D scan, resulting in a slower convergence speed of morphological deformation, and obvious residual errors and inaccurate correspondence of significant local parts in the final result.
To solve these problems, the author proposes a brand - new fully - automatic non - rigid 3D shape registration framework, which includes the following three main contributions:
1. **New 3D landmark location and pose normalization method**: Through an improved 3D landmark location and pose normalization method, ensure more accurate initial alignment between the template and the target.
2. **Adaptive shape template method**: By using sparse shape information (usually point landmarks) to obtain an adaptive template from the original template, the template can better match the specific features of the original 3D scan, thereby accelerating the convergence of the registration algorithm and achieving better final shape correspondence.
3. **New iterative registration method**: Combining the Iterative Closest Point (ICP) method and the Coherent Point Drift (CPD) method, a new iterative registration method - Iterative Coherent Point Drift (ICPD) is proposed to achieve more stable and accurate correspondence establishment than the standard CPD.
Through these improvements, the framework proposed by the author performs excellently in the qualitative and quantitative evaluations on three datasets (Headspace, BU3D, and the synthetic LSFM dataset), reaching the state - of - the - art performance.